九州大学学術情報リポジトリ
Kyushu University Institutional Repository
Labelling Method by Pupillometry for
Classifying Attention Level by EEG/ECG/NIRS
ゼニファ, ファディラ
https://doi.org/10.15017/4060012
出版情報：九州大学, 2019, 博士（システム生命科学）, 課程博士 バージョン：
権利関係：
（様式３）
氏 名 ：ゼニファ ファディラ
論 文 名 ：Labelling Method by Pupillometry for Classifying Attention Level by EEG-ECG-NIRS
(EEG・ECG・NIRS を用いた注意レベルの分類のための瞳孔測定によ るラベリング手法)
区 分 ：甲
論 文 内 容 の 要 旨
There are numerous methods to evaluate attention levels such as observation, self-assessment, and objective performance. This study aims to propose a new labeling method for attention levels detection by using parameter settings of pupillometry. This parameter setting then would be applied as data labeling in supervised machine learning toward EEG-ECG-NIRS.
To develop parameter settings of attention level evaluation, this study investigated the reaction of blink rates and pupillometry toward attention level based on self-assessment during cognitive tasks. My result showed there is no significant differences (P>0.05) in blink rates toward attention level within 10 seconds. On the other hand, pupillometry in low attention showed significant differences in pupillometry in the last 4 seconds cognitive tasks (P<0.05).
After that, I calculated the distribution fit of pupillometry reaction in the attention level of all participants and plot the critical point of pupillometry data in 10 seconds and 4 seconds. After doing several experimental procedures, I chose parameter setting with a percentage of error of less than 15% and a different error 35 % compare with self assesment as future labeling method.
Parameter setting which has been selected is when z-score within a specific range (-0.965 ≤ pupil ≤ 1.014) as high attention, other that range, will be classified as low attention.
Furthermore, I applied my labeling method for another physiological signal such as electroencephalograph (EEG), electrocardiograph (ECG), and near-infrared spectroscopy (NIRS). Numerous methods using electroencephalograph (EEG), electrocardiograph (ECG), and near-infrared spectroscopy (NIRS) for attention level detection have been proposed.
However, the results were either unsatisfactory or required many channels. In this study, I introduce the implementation of an EEG-ECG-NIRS for attention level detection. I used two-electrode wireless EEG, a wireless ECG, and two wireless channels NIRS to detect attention level during backward digit span, forward digit span and arithmetic. High attention will be labelled to data which has pupillometry z-score within specific range (-0.965 ≤ pupil ≤ 1.014) and another that range, will be classified as low attention. By using CFS+kNN algorithm, my result showed the accuracy system of EEG-ECG-NIRS (83.33± 5.95%) has the highest accuracy compare with EEG (81.90± 4.69%), ECG (82.51±3.57%), NIRS (78.37±7.12%).
Algorithm CFS+kNN also shown highest performance compare with other methods such as CFS+SVM (55.49± 27.89%), kNN (80.84± 3.88%) and SVM (55.88± 13.14%)
In summary, in this study, I established new parameter settings for evaluating attention level by using pupillometry and apply the parameter settings into EEG-ECG-NIRS to evaluate the EEG-ECG-NIRS performance, comparing with standalone system.
